Simluating a satellite connection using dummynet?

jdow jdow at earthlink.net
Sun Apr 23 23:03:10 UTC 2006


From: "Kevin Kinsey" <kdk at daleco.biz>

> Dan Busarow wrote:
> 
>>
>> On Apr 22, 2006, at 9:40 PM, Philip Hallstrom wrote:
>>
>>> Hi all -
>>>
>>> Odd question for you.  I have the opportunity to work
>>> from home,  but it would require using a sat internet
>>> connection (no cable or  dsl anywhere close).
>>>
>>> I've been reading up on it and best I can tell I'm looking
>>> at  1000ms round trips... at *best*.  Most of what I do
>>> I can do on  servers at home, but there will be the
>>> occasional ssh, etc.
>>>
>>> I recently setup ipfw/dummynet with a pipe and a 750ms
>>> delay both  in and out and it wasn't as bad as I thought
>>> it would be -- at  least for ssh/text.  Reminds me of my
>>> days on a 9600 baud modem. heh.
>>>
>>> I'm curious though whether this is a realistic test. 
>>> Thoughts?
>>>
>>> Any of you use satellite?  How do you find it?
>>
>>
>> I had StarBand for about two years.  1000ms RTT are the
>> best you will  see.  pushing 2000ms is more like it.
>>
>> While it is possible to work via an SSH session it will
>> try your  patience.
>>
>> It is doable, and it allowed me to move out to the
>> country, but  that's about it.  I now have a terrestrial
>> radio link into the  nearest town, 15 miles away, and
>> it's beautiful.
> 
> 
> My wife's employer had a Hughes connection for something
> over a year.  Generally speaking, they weren't impressed.
> 
> Their operation is a small insurance office, and they needed
> quick https service to the home office in Iowa, for a web app
> that seems to take a pretty quick pipe to operate well.
> 
> I never ran a sniffer, but it seemed as if their OS (Microsoft)
> had some trouble with this setup, particularly with https
> traffic.  You'd wait a good long time (few seconds), then get a
> big burst of data ... if you weren't cluttering things up with
> retries.  Since this HTTPS traffic was his "business", he
> decided it was more important to keep his employees
> happy, so he later decided to devote a portion of his
> disposable income to a local outfit that provides a
> T1 instead.  TCP/IP being as it is, it's likely that MSFT
> QoS was dropping the packet sizes to help deal
> with the "congestion" ;-), but I was never sure.
> 
> I'd concur that 1000 ms was a pretty normal RTT for
> ICMP, and it could, and often did go higher, a la
> 2500+.
> 
> Much like Dan, I use an 11Mbps LOS radio connection
> to the water tower about 4 miles away.  Nice, except
> I really need to raise my receiver so I can maintain
> good QoS when the foilage gets going....
> 
> I think grog@ has satellite service in AUS.  You might
> see if you can turn up anything on his site.
> 
> Kevin Kinsey

Four round trips of as much as 25,000 miles is a bit more than
half a second. The typically employed interleaving/deinterleaving
and error correction coding means you can easily add another half
a second to the path.

(Of course, for something like the data mode operation on Inmarsat-M,
which is 2400bps max, it was pretty bad. As part of testing the
implementation I ran an interesting test. Satellite from Torrance to
the Atlantic sat to Goonhilly England (the only ground station that
was useable for the testing) back to LAX by unknown path thence
tymnet to a service in Mass to connect to the Internet and then
connect back to itself via TCP/IP was a LONG SLOW interactive session.
Then I made a 1 megabyte file transfer. It worked. It was signed off.
And quite a few units presold with data mode option suddenly grew the
implementation, one of the first units to have it. Then I entered
burnout.... Anyway - by the time that was all connected turn around
varied from 2.5 to 5 seconds on typed characters.)

{o.o}   Joanne


More information about the freebsd-questions mailing list